Enhancing Reasoning Capabilities by Instruction Learning and Chain-of-Thoughts for Implicit Discourse Relation Recognition

Abstract

Implicit discourse relation recognition is a challenging task in natural language understanding that involves identifying the sense of connection between two arguments. This paper presents a novel approach that leverages generative models, instruction learning, and chain-of-thoughts to enhance reasoning capabilities for implicit discourse relation recognition. Our proposed method, IICOT, employs a combination of instruction templates and in-context learning to refine the generative model, and utilizes chain-of-thoughts to partition the inference process into a sequence of successive stages. Experimental results on benchmark datasets demonstrate the effectiveness of our approach, achieving state-of-the-art performance on all three datasets. Our analysis also reveals the benefits of incorporating explicit data and the denoising ability of chain-of-thoughts.

Key Contributions

• We propose a novel approach that leverages generative models, instruction learning, and chain-of-thoughts for implicit discourse relation recognition.
• Our method, IICOT, achieves state-of-the-art performance on benchmark datasets.
• We demonstrate the benefits of incorporating explicit data and the denoising ability of chain-of-thoughts.
• Our approach provides a new perspective on utilizing generative models for natural language understanding tasks.

Download Paper